Cross-Afọwọsi
Ifọwọsi-agbelebu jẹ ilana iṣatunṣe fun ṣiṣeroro bawo ni awoṣe yoo ṣe gbogbogbo si data ti a ko rii.
Akopọ
It makes better use of limited data and gives a more reliable performance estimate than a single train/test split.
Jin Dive
Pipin ọkọ oju-irin kan / pipin idanwo jẹ ẹlẹgẹ: Dimegilio ti o gba da lori pupọ lori iru awọn ori ila ti o ṣẹlẹ si ilẹ ni eto idanwo naa. Ifọwọsi-agbelebu ṣe atunṣe eyi nipa yiyi ipa ti ṣeto idanwo naa. Ni k-agbo agbelebu-afọwọsi, o pin awọn data sinu k dogba folds, reluwe lori k-1 ninu wọn, se ayẹwo lori awọn ti o waye-agbo, ki o si tun k igba ki gbogbo kana ni idanwo pato ẹẹkan. Apapọ awọn ikun k n mu iṣiro iduroṣinṣin diẹ sii pẹlu iwọn iyipada kan. Awọn aṣayan ti o wọpọ jẹ awọn ipada 5 tabi 10. Awọn iyatọ pẹlu stratified k-fold (titọju awọn iwọn kilasi fun data aiṣedeede), fi silẹ-ọkan-jade (k dọgba nọmba awọn ayẹwo), ati awọn ipin akoko-ila ti kii ṣe ikẹkọ ni ọjọ iwaju lati ṣe asọtẹlẹ ohun ti o ti kọja.
Imọ-imọ-ẹrọ
Ifọwọsi-agbelebu jẹ alagbara julọ fun yiyan awoṣe ati yiyi hyperparameter: o ṣe afiwe awọn atunto nipasẹ Dimegilio afọwọsi apapọ wọn ju ki o kọja si pipin kan. Ipalara to ṣe pataki ni jijo data - eyikeyi iṣaju ti o 'ri' gbogbo dataset (iwọn, yiyan ẹya, iṣiro) gbọdọ wa ni ibamu ninu agbo kọọkan, kii ṣe ṣaaju pipin, tabi iṣiro rẹ yoo jẹ abosi ireti. Ifọwọsi agbelebu ti itẹ-ẹiyẹ yapa iṣatunṣe lati igbelewọn ikẹhin lati yago fun jijo yii.
Ipa Ilana
Awọn ipinnu diẹ sii
O ṣe iranlọwọ fun ọ lati ya sọtọ awọn iṣeduro imọ-ẹrọ lati ede tita.
Iye owo ati isuna
O le beere awọn ibeere imuse to dara julọ ṣaaju lilo owo tabi akoko.
Ẹgbẹ ati ṣiṣan iṣẹ
Awọn ẹgbẹ pẹlu oye pinpin ṣe ọja to dara julọ, eto imulo, ati awọn ipinnu ikẹkọ.
Ojo iwaju ti Cross-Afọwọsi
Bi datasets ati awọn awoṣe dagba, nṣiṣẹ k ni kikun ikẹkọ waye di gbowolori, ki awọn oṣiṣẹ increasingly ojurere kan ti o tobi idaduro-jade afọwọsi ṣeto fun jin eko nigba ti ifiṣura agbelebu-afọwọsi fun kekere tabi tabular datasets. ML adaṣe ati awọn irinṣẹ bii scikit-learn's GridSearchCV ati Optuna ṣe afọwọsi agbelebu sinu wiwa hyperparameter nipasẹ aiyipada. Iwadi n tẹsiwaju lori awọn isunmọ ti o din owo, awọn opo gigun ti o le sọ jijo, ati afọwọsi to dara fun akojọpọ, ilana ati data ti o gbẹkẹle akoko.
Real-World imuse
Lilo 5-afọwọsi-agbelebu-afọwọsi lati ṣe afiwe ipadasẹhin logistic, igbo laileto, ati igbega gradient ṣaaju ṣiṣe si awoṣe kan.
Nbere k-agbo stratified lori dataset wiwa-jegudujera ti ko ni iwọntunwọnsi nitorina agbo kọọkan tọju aijọju iwọn-kilaasi toje kanna.
Nṣiṣẹ GridSearchCV tabi RandomizedSearchCV, eyiti o ṣeduro gbogbo apapọ hyperparameter lati mu awọn eto to dara julọ.
Lilo awọn ọna-akoko (yiyi / jijọ-siwaju) ijẹrisi-agbelebu lati ṣe iṣiro ọja iṣura tabi asọtẹlẹ eletan laisi ikẹkọ lori data iwaju.
Awọn ewu & Awọn ọna iṣọ
Awọn ẹgbẹ oriṣiriṣi le lo ọrọ kanna ni oriṣiriṣi, nitorinaa ṣalaye iwọn ni kutukutu.
Awọn aṣepari le wo lagbara lakoko ti iṣẹ-aye gidi ko ṣe deede.
Aibikita didara data ati awọn ero igbelewọn nigbagbogbo ṣẹda awọn abajade ẹlẹgẹ.
Ilana Ilana imuse
Bẹrẹ pẹlu itumọ-ede itele ti abajade ti o nilo.
Mu metiriki aṣeyọri kan ati ipo ikuna kan ṣaaju idanwo.
Ṣiṣe awakọ kekere kan pẹlu data aṣoju, kii ṣe eto demo didan.
Iwe-ipamọ nibiti Cross-Validation ṣe iranlọwọ ati nibiti awọn ọna ti o rọrun jẹ dara julọ.
Tesiwaju Ṣiṣawari
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Cross-Validation quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
Itọsọna atẹle
Cross-Encoders vs Bi-Encoders
Awọn ibeere ti a beere nigbagbogbo
What is Cross-Validation?
Ifọwọsi-agbelebu jẹ ilana iṣatunṣe fun ṣiṣeroro bawo ni awoṣe yoo ṣe gbogbogbo si data ti a ko rii. O ṣe lilo ti o dara julọ ti data to lopin ati funni ni iṣiro iṣẹ ṣiṣe ti o ni igbẹkẹle diẹ sii ju ọkọ oju-irin kan / pipin idanwo kan.
Ni boṣewa k-agbo agbelebu-afọwọsi, igba melo ni aaye data kọọkan lo fun idanwo?
Agbo kọọkan n ṣiṣẹ bi eto idanwo ni ẹẹkan kọja awọn iyipo k, nitorinaa gbogbo ayẹwo ni idanwo ni akoko kan ati ikẹkọ lori iyokù.
Kini anfani akọkọ ti k-agbo agbelebu-afọwọsi lori ọkọ oju-irin kan / pipin idanwo?
Nipa aropin lori ọpọ awọn agbo, afọwọsi agbelebu dinku iyatọ ti iṣiro iṣẹ ni akawe si gbigbekele pipin lainidii kan.
Kini idi ti k-agbo agbelebu-ifọwọsi ko yẹ fun asọtẹlẹ jara-akoko?
Yiyọ data akoko-paṣẹ n jo ojo iwaju sinu ikẹkọ, nitorinaa-akoko CV nlo awọn pipin-ipin-ipin ti o ṣe ikẹkọ nikan lori awọn akiyesi ti o kọja.